Juneseok Lee

Juneseok Lee

Professor of Civil & Environmental Engineering · Building water systems

iBWS: Intelligent Building Water Systems


iBWS — Intelligent Building Water Systems

Building water systems, the plumbing inside hospitals, hotels, schools, and homes, are linked to 8,000–18,000 Legionella-related hospitalizations annually in the U.S. and remain the least-monitored part of drinking water infrastructure. The phenomena that drive that risk are hard to observe directly.

iBWS is how I try to see them, understand them, and act on them.

  Question Tools
Seeing What is the water doing right now? Sensors, building information models, data
Understanding Why is it doing that? Physics-based hydraulic, transient, and thermal models; machine learning; physics-informed learning
Acting What should the building do next? Digital twins, multi-objective optimization, AI agents

Why this computation is necessary

The phenomena are too coupled to observe directly, too slow to simulate in real time, and too consequential to automate blindly.

  • High-performance computing — the physics are coupled and the operating space is enormous. A single building spans thousands of combinations of demand, temperature, occupancy, and season. That space has to be computed thoroughly, not sampled by hand.
  • Surrogate models — a coupled physics run takes minutes to hours, while optimization and real-time control need thousands to millions of evaluations. Surrogates trained on HPC results answer in milliseconds while carrying the physics behind them. HPC buys accuracy; surrogates buy speed.
  • Agent-based orchestration — a real decision, such as flushing a riser or lowering a setpoint, draws on several models, live sensor data, plumbing codes, and competing objectives at once, continuously. An agent coordinates those tools and explains its reasoning to the engineer responsible for the building.
Presenting the iBWS coupled components, 2025
Presenting the iBWS coupled components, Mobile, Alabama, 2025

Six Components

1 · BIM2WNTR — Automated Model Generation
Converts Building Information Models (BIM) to WNTR-ready hydraulic simulation networks automatically. Achieves R²>0.99 accuracy in under 10 seconds.

2 · Pump Scheduling Optimization
Multi-objective pump scheduling using NSGA-II and 54,000 HPC simulations. Achieves 45.6% energy savings relative to conventional always-on scheduling.

3 · Thermal Risk Modeling — Legionella Prevention
Physics-based thermal simulation identifying zones in the 25–45°C growth range, triggering automated flushing protocols before outbreak conditions develop.

4 · Hydraulic Transient Analysis
Pressure surge quantification across 96 building scenarios using TSNet. Identifies transient-induced pipe failure risks and informs protective control strategies.

5 · One Water Simulation
Coupled potable-drainage simulation enabling simultaneous modeling of water supply, greywater recycling, and drainage systems.

6 · iBWSS — Water Sustainability Optimization
Multi-objective optimization of rainwater harvesting, greywater recycling, and solar-assisted water heating. Achieves up to 62% water self-sufficiency across 12 U.S. cities.

Three Levels of Autonomous Control

What a building delivers goes directly into people’s bodies, so autonomy here means something narrower than “no people involved.” Engineers set the operating envelope, define what the system may do on its own, and review what it did. Within that envelope, routine actions run unattended; anything outside it returns to a person. Every automated action must be bounded, explainable, and reversible.

Level 1 Decision Support Agent diagnoses · Human executes Current · 2025–2026
Building sensors temp · flow · pressure iBWS agent LLM + physics models Decision report risk zones · work orders Facility manager reads report · decides · acts manual on-site action (flush valve, adjust pump, log incident) HUMAN IN LOOP
WHO ACTS: Facility manager — iBWS provides the diagnosis and ranked work order; a human performs every physical action manually on-site. Most existing buildings fall into this category today.
iBWS agent monitors sensor data and executes physics-based simulations (P06 thermal risk, P03 hydraulic model, P00 pump optimization). When a threshold is exceeded — e.g., Zone 3 at 35°C — the agent generates a prioritized action report. The facility manager reads the report and manually acts. No building automation hardware required. Deployable today.
Level 2 Semi-Autonomous Agent commands · System executes · Human oversees 3–5 years · BAS/BMS integration required
Sensors / IoT 24/7 monitoring iBWS agent LLM + physics models BAS / BMS system receives iBWS commands Smart valves · pumps auto-execute commands Human override / approval available at any step — operator can pause, override, or abort HUMAN OVERSIGHT
WHO ACTS: Automation executes iBWS commands through the building's BAS/BMS — human operator retains full override authority at any time. Requires smart valve and VFD pump infrastructure.
The iBWS agent issues commands directly to the Building Automation System (BAS) or Building Management System (BMS) via standard protocols (BACnet, Modbus). The BAS executes thermal flushing, pump scheduling adjustments, and setpoint changes automatically. Suitable for hospitals, hotels, and universities with existing smart building infrastructure. Reduces response time from hours to seconds.
Level 3 Fully Autonomous Agent monitors · decides · executes · reports — routine actions run unattended 10-year vision · 2035+
Smart sensors continuous data stream iBWS autonomous agent LLM + all validated models Smart infrastructure valves · pumps · thermal systems Audit log reports to stakeholders continuous 24/7 autonomous control loop routine decisions run without human intervention STAKEHOLDERS audit log only
WHO ACTS: The iBWS autonomous agent — monitors, decides, and executes routine water management actions within an envelope set by engineers. Conditions outside that envelope escalate to a person. Stakeholders receive audit reports.
Within its approved envelope, iBWS operates as a closed-loop system: real-time sensor data feeds continuously into the agent, which runs validated physics simulations, makes decisions using multi-objective optimization, and issues direct commands to smart infrastructure. Analogous to how HVAC building automation systems operate today — but for water. This is the iBWS 10-year research and commercialization target.
Sensors / data input
iBWS agent
Outputs / audit log
Automated execution
Human operator
Human override (optional)

Technology Stack

  • Simulation engines: WNTR, PySWMM, TSNet, custom thermal models
  • Optimization: NSGA-II, multi-objective Pareto optimization
  • AI orchestration: LLM agent frameworks, physics-informed neural surrogates
  • Data integration: BIM2WNTR automated pipeline
  • Computing: HPC-enabled simulation (NAIRR/ACCESS aligned)

Research Roadmap

Phase Timeline Milestone
Component validation 2002–2025 Six validated physics models
Agent architecture 2025–2026 Autonomous orchestration proof-of-concept
Field validation 2026–2028 Pilot deployment in real buildings
Generalized platform 2028–2030 Multi-building scalable product
Full autonomy 2030–2035 Level 3 autonomous control

Publications

Selected iBWS-related publications are listed on the Publications page. Key venues include ASCE Journal of Water Resources Planning and Management, AWWA Water Science, and ASCE/ASEE conference proceedings.

Collaboration & Contact

I welcome collaboration with researchers, building owners, utilities, and industry partners interested in intelligent building water systems.

Email: juneseok.lee@manhattan.edu
Google Scholar: View iBWS-related publications

iBWS builds on building water systems research since 2002.